Segmentation of Breast Cancer on Ultrasound Images using Attention U-Net Model

Abstract

Breast cancer (BC) is one of the most prevailing and life-threatening types of cancer impacting women worldwide. Early detection and accurate diagnosis are crucial for effective treatment and improved patient outcomes. Deep learning techniques have shown remarkable promise in medical image analysis tasks, particularly segmentation. This research leverages the Breast Ultrasound Images BUSI dataset to develop two variations of a segmentation model using the Attention U-Net architecture. In this study, we trained the Attention3 U-Net and the Attention4 U-net on the BUSI dataset, consisting of normal, benign, and malignant breast lesions. We evaluated the model's performance based on standard segmentation metrics such as the Dice coefficient and Intersection over Union (IoU). The results demonstrate the effectiveness of the Attention U-Net in accurately segmenting breast lesions, with high overall performance, indicating agreement between predicted and ground truth masks. The successful application of the Attention U-Net to the BUSI dataset holds promise for improving breast cancer diagnosis and treatment. It highlights the potential of deep learning in medical image analysis, paving the way for more efficient and reliable diagnostic tools in breast cancer management.